Understanding Annotator Safety Policy with Interpretability Continual Knowledge Updating in LLM Systems: Learning Through Multi-Timescale Memory Dynamics Dynamic Pondering Sparsity-aware Mixture-of-Experts Transformer for Event Stream based Visual Object Tracking Schedule-and-Calibrate: Utility-Guided Multi-Task Reinforcement Learning for Code LLMs Structured Progressive Knowledge Activation for LLM-Driven Neural Architecture Search